基于无人机多光谱影像的冬小麦花前期氮素营养估测  

Estimation of nitrogen nutrition before flowering stage of winter wheat based on UAV multispectral imagery

作　　者：陶新宇 朱永基 苏祥祥 祝雪晴 任兰天 刘吉凯 李新伟 TAO Xinyu;ZHU Yongji;SU Xiangxiang;ZHU Xueqing;REN Lantian;LIU Jikai;LI Xinwei(College of Resource and Environment,Anhui Science and Technology University,Fengyang 233100,China;Anhui Agricultural Waste Fertilizer Utilization and Cultivated Land Quality Improvement Engineering Research Center,Fengyang 233100,China;Anhui Crop Intelligent Planting and Processing Technology Engineering Research Center,Fengyang 233100,China)

机构地区：[1]安徽科技学院资源与环境学院,安徽凤阳233100 [2]安徽省农业废弃物肥料化利用与耕地质量提升工程研究中心,安徽凤阳233100 [3]安徽省作物智慧种植与加工技术工程研究中心,安徽凤阳233100

出　　处：《安徽科技学院学报》2023年第3期50-59,共10页Journal of Anhui Science and Technology University

基　　金：凤阳县科技计划项目(NY2022-01);新疆生产建设兵团绿洲生态重点实验室开放课题发展基金(201903);安徽省大学生创新创业训练计划项目(202210879043)。

摘　　要：目的:利用多光谱无人机建立长江中下游地区冬小麦花前期氮素营养低成本、易推广、高效率的估测模型。方法:利用DJI Phantom 4 Multispectral相机获取4个氮素水平下3个冬小麦品种的多光谱影像数据,探讨植被指数对4种冬小麦氮素营养参数估测的敏感性,采用线性回归、随机森林和主成分分析算法构建冬小麦花前关键生育时期的氮素营养参数监测模型,筛选各时期氮素营养参数的最优估测模型。结果:所选9种植被指数均与氮素营养参数呈极显著相关。在拔节期,线性模型对氮素营养参数的预测性能最佳,R^(2)为0.87~0.94,nRMSE为8.44~12.49,RPD为2.79~4.08;在孕穗期,3种模型的估测性能相当;在抽穗期,随机森林和主成分回归模型的估测精度更高。结论:线性回归模型在拔节期和孕穗期,随机森林和主成分回归模型在冠层结构复杂的抽穗期,可实现冬小麦氮素营养参数的精准监测,研究可为长江中下游冬小麦氮素营养诊断和施肥决策提供科学参考。Objective:To establish a low-cost,easy to popularize and efficient estimation model for nitrogen nutrition of winter wheat in the middle and lower reaches of the Yangtze River by using multi spectral UAVs.Methods:The DJI Phantom 4 Multispectral camera was used to obtain multispectral image data of three varieties of winter wheat under four nitrogen levels,and the sensitivity of vegetation index to the estimation of nitrogen nutrient parameters of four kinds of winter wheat was investigated.The component analysis algorithm constructed the monitoring model of nitrogen nutrition parameters in the key growth period before flowering of winter wheat,and screened the optimal estimation model of nitrogen nutrition parameters in each period.Results:The nine vegetation indices selected were all significantly correlated with nitrogen nutrition parameters;at the jointing stage,the linear model had the best prediction performance for the four nitrogen nutrition parameters,with R^(2)of 0.87—0.94 and nRMSE of 8.44—12.49,and the RPD ranged from 2.79 to 4.08.At the booting stage,the estimation performance of the three models was comparable.At the heading stage,the estimation accuracy of the random forest and principal component regression models was higher.Conclusion:The linear regression model can accurately monitor the nitrogen nutrition parameters of winter wheat at the jointing stage and booting stage,and the random forest and principal component regression model can realize the accurate monitoring of the nitrogen nutrition parameters of winter wheat at the heading stage with complex canopy structure.Decision-making fertilization provides scientific reference.

关 键 词：无人机多光谱影像 氮素营养参数 植被指数 花前期 冬小麦 

分 类 号：S127[农业科学—农业基础科学]